ML Well Production Prediction with Physics Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning methods for predicting well performance in unconventional reservoirs face challenges with small training data sets, leading to suboptimal results and a need for incorporating domain knowledge, such as physics constraints, to improve model training.
Innovation Solution
The method involves generating multiple artificial neural network models based on initial parameters, selecting top-ranked models using loss values from both training and validation data sets, and constraining them with physical rules like the perforated well length rule to generate final predicted well production data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning method is used for predicting well performance, then prediction capability is improved, but the requirement for large training data set increases
Solution Approach 1:
The patent changes the approach from using raw data quantity to using data quality and physics-based parameter transformations. By constraining model parameters with physical laws (e.g., dimensionless groups, causality relationships), the method achieves reliable predictions with small data sets by transforming the problem from data-hungry to physics-guided parameter estimation.
Solution Approach 2:
The patent introduces physics-based constraints as an intermediary between the machine learning model and the training data. These constraints act as a mediator that guides the model to find physically meaningful relationships, allowing the model to generalize from small data sets by incorporating domain knowledge through physical principles rather than relying solely on data volume.
2Reliability
If more models are trained to improve prediction accuracy, then prediction reliability is improved, but computational complexity increases
Solution Approach 1:
The patent segments the model training process into multiple individually trained models with different initial parameters, then applies physics constraints to filter and select the most appropriate models. This segmentation allows parallel training of multiple models while using physics-based filtering to reduce the final set to only those that satisfy physical laws, balancing accuracy with computational efficiency.
Solution Approach 2:
The patent implements a feedback mechanism where model predictions are continuously evaluated against physical constraints and validation data. Models that violate physical laws or perform poorly on validation data are eliminated, and the process iterates until a set of physically consistent models remains. This feedback loop ensures high prediction accuracy while maintaining computational efficiency by eliminating invalid models early.
Data Source
AI summary
A method and a system for predicting well production of a reservoir using machine learning models and algorithms is disclosed. The method includes obtaining a training data set for training a machine learning (ML) model and selecting an artificial neural network model structure, the model structure including a number of layers and a number of nodes of each layer. Further, the method includes generating a plurality of individually trained ML models and calculating a model performance of each trained model by evaluating a difference between a model prediction and a well performance data. The plurality of top-ranked individually trained ML models is constrained using one or multiple known physical rules. A plurality of individual predicted well production data is generated using the geological, the completion, and the petrophysical data of interest and a final predicted well production data is generating based on the plurality of individual predicted well production data.